1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2017 Intel Corporation
5 #include <rte_atomic.h>
6 #include <rte_branch_prediction.h>
7 #include <rte_byteorder.h>
8 #include <rte_common.h>
10 #include <rte_ethdev.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_malloc.h>
13 #include <rte_bus_vdev.h>
14 #include <rte_kvargs.h>
16 #include <rte_debug.h>
19 #include <sys/types.h>
21 #include <sys/socket.h>
22 #include <sys/ioctl.h>
23 #include <sys/utsname.h>
30 #include <arpa/inet.h>
32 #include <linux/if_tun.h>
33 #include <linux/if_ether.h>
34 #include <linux/version.h>
37 #include <rte_eth_tap.h>
39 #include <tap_netlink.h>
40 #include <tap_tcmsgs.h>
42 /* Linux based path to the TUN device */
43 #define TUN_TAP_DEV_PATH "/dev/net/tun"
44 #define DEFAULT_TAP_NAME "dtap"
46 #define ETH_TAP_IFACE_ARG "iface"
47 #define ETH_TAP_SPEED_ARG "speed"
48 #define ETH_TAP_REMOTE_ARG "remote"
49 #define ETH_TAP_MAC_ARG "mac"
50 #define ETH_TAP_MAC_FIXED "fixed"
52 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
53 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
55 static struct rte_vdev_driver pmd_tap_drv;
57 static const char *valid_arguments[] = {
67 static volatile uint32_t tap_trigger; /* Rx trigger */
69 static struct rte_eth_link pmd_link = {
70 .link_speed = ETH_SPEED_NUM_10G,
71 .link_duplex = ETH_LINK_FULL_DUPLEX,
72 .link_status = ETH_LINK_DOWN,
73 .link_autoneg = ETH_LINK_SPEED_AUTONEG
77 tap_trigger_cb(int sig __rte_unused)
79 /* Valid trigger values are nonzero */
80 tap_trigger = (tap_trigger + 1) | 0x80000000;
83 /* Specifies on what netdevices the ioctl should be applied */
90 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
92 /* Tun/Tap allocation routine
94 * name is the number of the interface to use, unless NULL to take the host
98 tun_alloc(struct pmd_internals *pmd)
101 #ifdef IFF_MULTI_QUEUE
102 unsigned int features;
106 memset(&ifr, 0, sizeof(struct ifreq));
109 * Do not set IFF_NO_PI as packet information header will be needed
110 * to check if a received packet has been truncated.
112 ifr.ifr_flags = IFF_TAP;
113 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
115 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
117 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
119 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
123 #ifdef IFF_MULTI_QUEUE
124 /* Grab the TUN features to verify we can work multi-queue */
125 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
126 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
129 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
131 if (features & IFF_MULTI_QUEUE) {
132 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
133 RTE_PMD_TAP_MAX_QUEUES);
134 ifr.ifr_flags |= IFF_MULTI_QUEUE;
138 ifr.ifr_flags |= IFF_ONE_QUEUE;
139 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
142 /* Set the TUN/TAP configuration and set the name if needed */
143 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
144 RTE_LOG(WARNING, PMD,
145 "Unable to set TUNSETIFF for %s\n",
151 /* Always set the file descriptor to non-blocking */
152 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
153 RTE_LOG(WARNING, PMD,
154 "Unable to set %s to nonblocking\n",
156 perror("F_SETFL, NONBLOCK");
160 /* Set up trigger to optimize empty Rx bursts */
164 int flags = fcntl(fd, F_GETFL);
166 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
168 if (sa.sa_handler != tap_trigger_cb) {
170 * Make sure SIGIO is not already taken. This is done
171 * as late as possible to leave the application a
172 * chance to set up its own signal handler first.
174 if (sa.sa_handler != SIG_IGN &&
175 sa.sa_handler != SIG_DFL) {
179 sa = (struct sigaction){
180 .sa_flags = SA_RESTART,
181 .sa_handler = tap_trigger_cb,
183 if (sigaction(SIGIO, &sa, NULL) == -1)
186 /* Enable SIGIO on file descriptor */
187 fcntl(fd, F_SETFL, flags | O_ASYNC);
188 fcntl(fd, F_SETOWN, getpid());
191 /* Disable trigger globally in case of error */
193 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
206 tap_verify_csum(struct rte_mbuf *mbuf)
208 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
209 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
210 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
211 unsigned int l2_len = sizeof(struct ether_hdr);
217 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
219 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
221 /* Don't verify checksum for packets with discontinuous L2 header */
222 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
223 rte_pktmbuf_data_len(mbuf)))
225 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
226 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
227 struct ipv4_hdr *iph = l3_hdr;
229 /* ihl contains the number of 4-byte words in the header */
230 l3_len = 4 * (iph->version_ihl & 0xf);
231 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
234 cksum = ~rte_raw_cksum(iph, l3_len);
235 mbuf->ol_flags |= cksum ?
236 PKT_RX_IP_CKSUM_BAD :
237 PKT_RX_IP_CKSUM_GOOD;
238 } else if (l3 == RTE_PTYPE_L3_IPV6) {
239 l3_len = sizeof(struct ipv6_hdr);
241 /* IPv6 extensions are not supported */
244 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
245 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
246 /* Don't verify checksum for multi-segment packets. */
247 if (mbuf->nb_segs > 1)
249 if (l3 == RTE_PTYPE_L3_IPV4)
250 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
251 else if (l3 == RTE_PTYPE_L3_IPV6)
252 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
253 mbuf->ol_flags |= cksum ?
254 PKT_RX_L4_CKSUM_BAD :
255 PKT_RX_L4_CKSUM_GOOD;
259 /* Callback to handle the rx burst of packets to the correct interface and
260 * file descriptor(s) in a multi-queue setup.
263 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
265 struct rx_queue *rxq = queue;
267 unsigned long num_rx_bytes = 0;
268 uint32_t trigger = tap_trigger;
270 if (trigger == rxq->trigger_seen)
273 rxq->trigger_seen = trigger;
274 rte_compiler_barrier();
275 for (num_rx = 0; num_rx < nb_pkts; ) {
276 struct rte_mbuf *mbuf = rxq->pool;
277 struct rte_mbuf *seg = NULL;
278 struct rte_mbuf *new_tail = NULL;
279 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
282 len = readv(rxq->fd, *rxq->iovecs,
283 1 + (rxq->rxmode->enable_scatter ?
284 rxq->nb_rx_desc : 1));
285 if (len < (int)sizeof(struct tun_pi))
288 /* Packet couldn't fit in the provided mbuf */
289 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
290 rxq->stats.ierrors++;
294 len -= sizeof(struct tun_pi);
297 mbuf->port = rxq->in_port;
299 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
301 if (unlikely(!buf)) {
302 rxq->stats.rx_nombuf++;
303 /* No new buf has been allocated: do nothing */
304 if (!new_tail || !seg)
308 rte_pktmbuf_free(mbuf);
312 seg = seg ? seg->next : mbuf;
313 if (rxq->pool == mbuf)
316 new_tail->next = buf;
318 new_tail->next = seg->next;
320 /* iovecs[0] is reserved for packet info (pi) */
321 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
322 buf->buf_len - data_off;
323 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
324 (char *)buf->buf_addr + data_off;
326 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
327 seg->data_off = data_off;
329 len -= seg->data_len;
333 /* First segment has headroom, not the others */
337 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
339 if (rxq->rxmode->hw_ip_checksum)
340 tap_verify_csum(mbuf);
342 /* account for the receive frame */
343 bufs[num_rx++] = mbuf;
344 num_rx_bytes += mbuf->pkt_len;
347 rxq->stats.ipackets += num_rx;
348 rxq->stats.ibytes += num_rx_bytes;
354 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
357 void *l3_hdr = packet + l2_len;
359 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
360 struct ipv4_hdr *iph = l3_hdr;
363 iph->hdr_checksum = 0;
364 cksum = rte_raw_cksum(iph, l3_len);
365 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
367 if (ol_flags & PKT_TX_L4_MASK) {
373 l4_hdr = packet + l2_len + l3_len;
374 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
375 l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
376 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
377 l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
381 if (ol_flags & PKT_TX_IPV4) {
382 struct ipv4_hdr *iph = l3_hdr;
384 l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
385 cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
387 struct ipv6_hdr *ip6h = l3_hdr;
389 /* payload_len does not include ext headers */
390 l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
391 l3_len + sizeof(struct ipv6_hdr);
392 cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
394 cksum += rte_raw_cksum(l4_hdr, l4_len);
395 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
396 cksum = (~cksum) & 0xffff;
403 /* Callback to handle sending packets from the tap interface
406 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
408 struct tx_queue *txq = queue;
410 unsigned long num_tx_bytes = 0;
414 if (unlikely(nb_pkts == 0))
417 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
418 for (i = 0; i < nb_pkts; i++) {
419 struct rte_mbuf *mbuf = bufs[num_tx];
420 struct iovec iovecs[mbuf->nb_segs + 1];
421 struct tun_pi pi = { .flags = 0 };
422 struct rte_mbuf *seg = mbuf;
423 char m_copy[mbuf->data_len];
427 /* stats.errs will be incremented */
428 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
431 iovecs[0].iov_base = π
432 iovecs[0].iov_len = sizeof(pi);
433 for (j = 1; j <= mbuf->nb_segs; j++) {
434 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
436 rte_pktmbuf_mtod(seg, void *);
439 if (mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
440 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
441 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) {
442 /* Support only packets with all data in the same seg */
443 if (mbuf->nb_segs > 1)
445 /* To change checksums, work on a copy of data. */
446 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
447 rte_pktmbuf_data_len(mbuf));
448 tap_tx_offload(m_copy, mbuf->ol_flags,
449 mbuf->l2_len, mbuf->l3_len);
450 iovecs[1].iov_base = m_copy;
452 /* copy the tx frame data */
453 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
458 num_tx_bytes += mbuf->pkt_len;
459 rte_pktmbuf_free(mbuf);
462 txq->stats.opackets += num_tx;
463 txq->stats.errs += nb_pkts - num_tx;
464 txq->stats.obytes += num_tx_bytes;
470 tap_ioctl_req2str(unsigned long request)
474 return "SIOCSIFFLAGS";
476 return "SIOCGIFFLAGS";
478 return "SIOCGIFHWADDR";
480 return "SIOCSIFHWADDR";
488 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
489 struct ifreq *ifr, int set, enum ioctl_mode mode)
491 short req_flags = ifr->ifr_flags;
492 int remote = pmd->remote_if_index &&
493 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
495 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
498 * If there is a remote netdevice, apply ioctl on it, then apply it on
503 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
504 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
505 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
508 /* fetch current flags to leave other flags untouched */
509 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
512 ifr->ifr_flags |= req_flags;
514 ifr->ifr_flags &= ~req_flags;
522 RTE_ASSERT(!"unsupported request type: must not happen");
524 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
526 if (remote-- && mode == LOCAL_AND_REMOTE)
531 RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
532 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
537 tap_link_set_down(struct rte_eth_dev *dev)
539 struct pmd_internals *pmd = dev->data->dev_private;
540 struct ifreq ifr = { .ifr_flags = IFF_UP };
542 dev->data->dev_link.link_status = ETH_LINK_DOWN;
543 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
547 tap_link_set_up(struct rte_eth_dev *dev)
549 struct pmd_internals *pmd = dev->data->dev_private;
550 struct ifreq ifr = { .ifr_flags = IFF_UP };
552 dev->data->dev_link.link_status = ETH_LINK_UP;
553 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
557 tap_dev_start(struct rte_eth_dev *dev)
561 err = tap_intr_handle_set(dev, 1);
564 return tap_link_set_up(dev);
567 /* This function gets called when the current port gets stopped.
570 tap_dev_stop(struct rte_eth_dev *dev)
572 tap_intr_handle_set(dev, 0);
573 tap_link_set_down(dev);
577 tap_dev_configure(struct rte_eth_dev *dev)
579 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
581 "%s: number of rx queues %d exceeds max num of queues %d\n",
583 dev->data->nb_rx_queues,
584 RTE_PMD_TAP_MAX_QUEUES);
587 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
589 "%s: number of tx queues %d exceeds max num of queues %d\n",
591 dev->data->nb_tx_queues,
592 RTE_PMD_TAP_MAX_QUEUES);
596 RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
597 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
599 RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
600 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
606 tap_dev_speed_capa(void)
608 uint32_t speed = pmd_link.link_speed;
611 if (speed >= ETH_SPEED_NUM_10M)
612 capa |= ETH_LINK_SPEED_10M;
613 if (speed >= ETH_SPEED_NUM_100M)
614 capa |= ETH_LINK_SPEED_100M;
615 if (speed >= ETH_SPEED_NUM_1G)
616 capa |= ETH_LINK_SPEED_1G;
617 if (speed >= ETH_SPEED_NUM_5G)
618 capa |= ETH_LINK_SPEED_2_5G;
619 if (speed >= ETH_SPEED_NUM_5G)
620 capa |= ETH_LINK_SPEED_5G;
621 if (speed >= ETH_SPEED_NUM_10G)
622 capa |= ETH_LINK_SPEED_10G;
623 if (speed >= ETH_SPEED_NUM_20G)
624 capa |= ETH_LINK_SPEED_20G;
625 if (speed >= ETH_SPEED_NUM_25G)
626 capa |= ETH_LINK_SPEED_25G;
627 if (speed >= ETH_SPEED_NUM_40G)
628 capa |= ETH_LINK_SPEED_40G;
629 if (speed >= ETH_SPEED_NUM_50G)
630 capa |= ETH_LINK_SPEED_50G;
631 if (speed >= ETH_SPEED_NUM_56G)
632 capa |= ETH_LINK_SPEED_56G;
633 if (speed >= ETH_SPEED_NUM_100G)
634 capa |= ETH_LINK_SPEED_100G;
640 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
642 struct pmd_internals *internals = dev->data->dev_private;
644 dev_info->if_index = internals->if_index;
645 dev_info->max_mac_addrs = 1;
646 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
647 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
648 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
649 dev_info->min_rx_bufsize = 0;
650 dev_info->pci_dev = NULL;
651 dev_info->speed_capa = tap_dev_speed_capa();
652 dev_info->rx_offload_capa = (DEV_RX_OFFLOAD_IPV4_CKSUM |
653 DEV_RX_OFFLOAD_UDP_CKSUM |
654 DEV_RX_OFFLOAD_TCP_CKSUM);
655 dev_info->tx_offload_capa =
656 (DEV_TX_OFFLOAD_IPV4_CKSUM |
657 DEV_TX_OFFLOAD_UDP_CKSUM |
658 DEV_TX_OFFLOAD_TCP_CKSUM);
662 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
664 unsigned int i, imax;
665 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
666 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
667 unsigned long rx_nombuf = 0, ierrors = 0;
668 const struct pmd_internals *pmd = dev->data->dev_private;
670 /* rx queue statistics */
671 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
672 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
673 for (i = 0; i < imax; i++) {
674 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
675 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
676 rx_total += tap_stats->q_ipackets[i];
677 rx_bytes_total += tap_stats->q_ibytes[i];
678 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
679 ierrors += pmd->rxq[i].stats.ierrors;
682 /* tx queue statistics */
683 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
684 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
686 for (i = 0; i < imax; i++) {
687 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
688 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
689 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
690 tx_total += tap_stats->q_opackets[i];
691 tx_err_total += tap_stats->q_errors[i];
692 tx_bytes_total += tap_stats->q_obytes[i];
695 tap_stats->ipackets = rx_total;
696 tap_stats->ibytes = rx_bytes_total;
697 tap_stats->ierrors = ierrors;
698 tap_stats->rx_nombuf = rx_nombuf;
699 tap_stats->opackets = tx_total;
700 tap_stats->oerrors = tx_err_total;
701 tap_stats->obytes = tx_bytes_total;
706 tap_stats_reset(struct rte_eth_dev *dev)
709 struct pmd_internals *pmd = dev->data->dev_private;
711 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
712 pmd->rxq[i].stats.ipackets = 0;
713 pmd->rxq[i].stats.ibytes = 0;
714 pmd->rxq[i].stats.ierrors = 0;
715 pmd->rxq[i].stats.rx_nombuf = 0;
717 pmd->txq[i].stats.opackets = 0;
718 pmd->txq[i].stats.errs = 0;
719 pmd->txq[i].stats.obytes = 0;
724 tap_dev_close(struct rte_eth_dev *dev)
727 struct pmd_internals *internals = dev->data->dev_private;
729 tap_link_set_down(dev);
730 tap_flow_flush(dev, NULL);
731 tap_flow_implicit_flush(internals, NULL);
733 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
734 if (internals->rxq[i].fd != -1) {
735 close(internals->rxq[i].fd);
736 internals->rxq[i].fd = -1;
738 if (internals->txq[i].fd != -1) {
739 close(internals->txq[i].fd);
740 internals->txq[i].fd = -1;
744 if (internals->remote_if_index) {
745 /* Restore initial remote state */
746 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
747 &internals->remote_initial_flags);
752 tap_rx_queue_release(void *queue)
754 struct rx_queue *rxq = queue;
756 if (rxq && (rxq->fd > 0)) {
759 rte_pktmbuf_free(rxq->pool);
760 rte_free(rxq->iovecs);
767 tap_tx_queue_release(void *queue)
769 struct tx_queue *txq = queue;
771 if (txq && (txq->fd > 0)) {
778 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
780 struct rte_eth_link *dev_link = &dev->data->dev_link;
781 struct pmd_internals *pmd = dev->data->dev_private;
782 struct ifreq ifr = { .ifr_flags = 0 };
784 if (pmd->remote_if_index) {
785 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
786 if (!(ifr.ifr_flags & IFF_UP) ||
787 !(ifr.ifr_flags & IFF_RUNNING)) {
788 dev_link->link_status = ETH_LINK_DOWN;
792 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
793 dev_link->link_status =
794 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
801 tap_promisc_enable(struct rte_eth_dev *dev)
803 struct pmd_internals *pmd = dev->data->dev_private;
804 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
806 dev->data->promiscuous = 1;
807 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
808 if (pmd->remote_if_index && !pmd->flow_isolate)
809 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
813 tap_promisc_disable(struct rte_eth_dev *dev)
815 struct pmd_internals *pmd = dev->data->dev_private;
816 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
818 dev->data->promiscuous = 0;
819 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
820 if (pmd->remote_if_index && !pmd->flow_isolate)
821 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
825 tap_allmulti_enable(struct rte_eth_dev *dev)
827 struct pmd_internals *pmd = dev->data->dev_private;
828 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
830 dev->data->all_multicast = 1;
831 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
832 if (pmd->remote_if_index && !pmd->flow_isolate)
833 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
837 tap_allmulti_disable(struct rte_eth_dev *dev)
839 struct pmd_internals *pmd = dev->data->dev_private;
840 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
842 dev->data->all_multicast = 0;
843 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
844 if (pmd->remote_if_index && !pmd->flow_isolate)
845 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
849 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
851 struct pmd_internals *pmd = dev->data->dev_private;
852 enum ioctl_mode mode = LOCAL_ONLY;
855 if (is_zero_ether_addr(mac_addr)) {
856 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
860 /* Check the actual current MAC address on the tap netdevice */
861 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
863 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
866 /* Check the current MAC address on the remote */
867 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
869 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
871 mode = LOCAL_AND_REMOTE;
872 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
873 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
874 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
876 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
877 if (pmd->remote_if_index && !pmd->flow_isolate) {
878 /* Replace MAC redirection rule after a MAC change */
879 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
881 "%s: Couldn't delete MAC redirection rule\n",
885 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
887 "%s: Couldn't add MAC redirection rule\n",
893 tap_setup_queue(struct rte_eth_dev *dev,
894 struct pmd_internals *internals,
901 struct pmd_internals *pmd = dev->data->dev_private;
902 struct rx_queue *rx = &internals->rxq[qid];
903 struct tx_queue *tx = &internals->txq[qid];
915 /* fd for this queue already exists */
916 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
917 pmd->name, *fd, dir, qid);
918 } else if (*other_fd != -1) {
919 /* Only other_fd exists. dup it */
920 *fd = dup(*other_fd);
923 RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
927 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
928 pmd->name, *other_fd, dir, qid, *fd);
930 /* Both RX and TX fds do not exist (equal -1). Create fd */
931 *fd = tun_alloc(pmd);
933 *fd = -1; /* restore original value */
934 RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
938 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
939 pmd->name, dir, qid, *fd);
942 tx->mtu = &dev->data->mtu;
943 rx->rxmode = &dev->data->dev_conf.rxmode;
949 tap_rx_queue_setup(struct rte_eth_dev *dev,
950 uint16_t rx_queue_id,
952 unsigned int socket_id,
953 const struct rte_eth_rxconf *rx_conf __rte_unused,
954 struct rte_mempool *mp)
956 struct pmd_internals *internals = dev->data->dev_private;
957 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
958 struct rte_mbuf **tmp = &rxq->pool;
959 long iov_max = sysconf(_SC_IOV_MAX);
960 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
961 struct iovec (*iovecs)[nb_desc + 1];
962 int data_off = RTE_PKTMBUF_HEADROOM;
967 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
968 RTE_LOG(WARNING, PMD,
969 "nb_rx_queues %d too small or mempool NULL\n",
970 dev->data->nb_rx_queues);
975 rxq->trigger_seen = 1; /* force initial burst */
976 rxq->in_port = dev->data->port_id;
977 rxq->nb_rx_desc = nb_desc;
978 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
981 RTE_LOG(WARNING, PMD,
982 "%s: Couldn't allocate %d RX descriptors\n",
983 dev->device->name, nb_desc);
986 rxq->iovecs = iovecs;
988 dev->data->rx_queues[rx_queue_id] = rxq;
989 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
995 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
996 (*rxq->iovecs)[0].iov_base = &rxq->pi;
998 for (i = 1; i <= nb_desc; i++) {
999 *tmp = rte_pktmbuf_alloc(rxq->mp);
1001 RTE_LOG(WARNING, PMD,
1002 "%s: couldn't allocate memory for queue %d\n",
1003 dev->device->name, rx_queue_id);
1007 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1008 (*rxq->iovecs)[i].iov_base =
1009 (char *)(*tmp)->buf_addr + data_off;
1011 tmp = &(*tmp)->next;
1014 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
1015 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1020 rte_pktmbuf_free(rxq->pool);
1022 rte_free(rxq->iovecs);
1028 tap_tx_queue_setup(struct rte_eth_dev *dev,
1029 uint16_t tx_queue_id,
1030 uint16_t nb_tx_desc __rte_unused,
1031 unsigned int socket_id __rte_unused,
1032 const struct rte_eth_txconf *tx_conf __rte_unused)
1034 struct pmd_internals *internals = dev->data->dev_private;
1037 if (tx_queue_id >= dev->data->nb_tx_queues)
1040 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1041 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1045 RTE_LOG(DEBUG, PMD, " TX TAP device name %s, qid %d on fd %d\n",
1046 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
1052 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1054 struct pmd_internals *pmd = dev->data->dev_private;
1055 struct ifreq ifr = { .ifr_mtu = mtu };
1058 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1060 dev->data->mtu = mtu;
1066 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1067 struct ether_addr *mc_addr_set __rte_unused,
1068 uint32_t nb_mc_addr __rte_unused)
1071 * Nothing to do actually: the tap has no filtering whatsoever, every
1072 * packet is received.
1078 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1080 struct rte_eth_dev *dev = arg;
1081 struct pmd_internals *pmd = dev->data->dev_private;
1082 struct ifinfomsg *info = NLMSG_DATA(nh);
1084 if (nh->nlmsg_type != RTM_NEWLINK ||
1085 (info->ifi_index != pmd->if_index &&
1086 info->ifi_index != pmd->remote_if_index))
1088 return tap_link_update(dev, 0);
1092 tap_dev_intr_handler(void *cb_arg)
1094 struct rte_eth_dev *dev = cb_arg;
1095 struct pmd_internals *pmd = dev->data->dev_private;
1097 nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1101 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1103 struct pmd_internals *pmd = dev->data->dev_private;
1105 /* In any case, disable interrupt if the conf is no longer there. */
1106 if (!dev->data->dev_conf.intr_conf.lsc) {
1107 if (pmd->intr_handle.fd != -1) {
1108 nl_final(pmd->intr_handle.fd);
1109 rte_intr_callback_unregister(&pmd->intr_handle,
1110 tap_dev_intr_handler, dev);
1115 pmd->intr_handle.fd = nl_init(RTMGRP_LINK);
1116 if (unlikely(pmd->intr_handle.fd == -1))
1118 return rte_intr_callback_register(
1119 &pmd->intr_handle, tap_dev_intr_handler, dev);
1121 nl_final(pmd->intr_handle.fd);
1122 return rte_intr_callback_unregister(&pmd->intr_handle,
1123 tap_dev_intr_handler, dev);
1126 static const uint32_t*
1127 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1129 static const uint32_t ptypes[] = {
1130 RTE_PTYPE_INNER_L2_ETHER,
1131 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1132 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1133 RTE_PTYPE_INNER_L3_IPV4,
1134 RTE_PTYPE_INNER_L3_IPV4_EXT,
1135 RTE_PTYPE_INNER_L3_IPV6,
1136 RTE_PTYPE_INNER_L3_IPV6_EXT,
1137 RTE_PTYPE_INNER_L4_FRAG,
1138 RTE_PTYPE_INNER_L4_UDP,
1139 RTE_PTYPE_INNER_L4_TCP,
1140 RTE_PTYPE_INNER_L4_SCTP,
1142 RTE_PTYPE_L2_ETHER_VLAN,
1143 RTE_PTYPE_L2_ETHER_QINQ,
1145 RTE_PTYPE_L3_IPV4_EXT,
1146 RTE_PTYPE_L3_IPV6_EXT,
1158 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1159 struct rte_eth_fc_conf *fc_conf)
1161 fc_conf->mode = RTE_FC_NONE;
1166 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1167 struct rte_eth_fc_conf *fc_conf)
1169 if (fc_conf->mode != RTE_FC_NONE)
1174 static const struct eth_dev_ops ops = {
1175 .dev_start = tap_dev_start,
1176 .dev_stop = tap_dev_stop,
1177 .dev_close = tap_dev_close,
1178 .dev_configure = tap_dev_configure,
1179 .dev_infos_get = tap_dev_info,
1180 .rx_queue_setup = tap_rx_queue_setup,
1181 .tx_queue_setup = tap_tx_queue_setup,
1182 .rx_queue_release = tap_rx_queue_release,
1183 .tx_queue_release = tap_tx_queue_release,
1184 .flow_ctrl_get = tap_flow_ctrl_get,
1185 .flow_ctrl_set = tap_flow_ctrl_set,
1186 .link_update = tap_link_update,
1187 .dev_set_link_up = tap_link_set_up,
1188 .dev_set_link_down = tap_link_set_down,
1189 .promiscuous_enable = tap_promisc_enable,
1190 .promiscuous_disable = tap_promisc_disable,
1191 .allmulticast_enable = tap_allmulti_enable,
1192 .allmulticast_disable = tap_allmulti_disable,
1193 .mac_addr_set = tap_mac_set,
1194 .mtu_set = tap_mtu_set,
1195 .set_mc_addr_list = tap_set_mc_addr_list,
1196 .stats_get = tap_stats_get,
1197 .stats_reset = tap_stats_reset,
1198 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1199 .filter_ctrl = tap_dev_filter_ctrl,
1203 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1204 char *remote_iface, int fixed_mac_type)
1206 int numa_node = rte_socket_id();
1207 struct rte_eth_dev *dev;
1208 struct pmd_internals *pmd;
1209 struct rte_eth_dev_data *data;
1213 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1215 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1217 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1221 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1223 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1227 pmd = dev->data->dev_private;
1229 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1231 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1232 if (pmd->ioctl_sock == -1) {
1234 "TAP Unable to get a socket for management: %s\n",
1239 /* Setup some default values */
1240 rte_memcpy(data, dev->data, sizeof(*data));
1241 data->dev_private = pmd;
1242 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1243 data->numa_node = numa_node;
1245 data->dev_link = pmd_link;
1246 data->mac_addrs = &pmd->eth_addr;
1247 /* Set the number of RX and TX queues */
1248 data->nb_rx_queues = 0;
1249 data->nb_tx_queues = 0;
1252 dev->dev_ops = &ops;
1253 dev->rx_pkt_burst = pmd_rx_burst;
1254 dev->tx_pkt_burst = pmd_tx_burst;
1256 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1257 pmd->intr_handle.fd = -1;
1259 /* Presetup the fds to -1 as being not valid */
1260 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1261 pmd->rxq[i].fd = -1;
1262 pmd->txq[i].fd = -1;
1265 if (fixed_mac_type) {
1266 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1267 static int iface_idx;
1268 char mac[ETHER_ADDR_LEN] = "\0dtap";
1270 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1271 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1273 eth_random_addr((uint8_t *)&pmd->eth_addr);
1276 /* Immediately create the netdevice (this will create the 1st queue). */
1278 if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1281 if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1284 ifr.ifr_mtu = dev->data->mtu;
1285 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1288 memset(&ifr, 0, sizeof(struct ifreq));
1289 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1290 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1291 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1295 * Set up everything related to rte_flow:
1297 * - tap / remote if_index
1298 * - mandatory QDISCs
1299 * - rte_flow actual/implicit lists
1302 pmd->nlsk_fd = nl_init(0);
1303 if (pmd->nlsk_fd == -1) {
1304 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1306 goto disable_rte_flow;
1308 pmd->if_index = if_nametoindex(pmd->name);
1309 if (!pmd->if_index) {
1310 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1311 goto disable_rte_flow;
1313 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1314 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1316 goto disable_rte_flow;
1318 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1319 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1321 goto disable_rte_flow;
1323 LIST_INIT(&pmd->flows);
1325 if (strlen(remote_iface)) {
1326 pmd->remote_if_index = if_nametoindex(remote_iface);
1327 if (!pmd->remote_if_index) {
1328 RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1329 pmd->name, remote_iface);
1332 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1333 "%s", remote_iface);
1335 /* Save state of remote device */
1336 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1338 /* Replicate remote MAC address */
1339 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1340 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1341 pmd->name, pmd->remote_iface);
1344 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1346 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1347 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1348 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1349 pmd->name, remote_iface);
1354 * Flush usually returns negative value because it tries to
1355 * delete every QDISC (and on a running device, one QDISC at
1356 * least is needed). Ignore negative return value.
1358 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1359 if (qdisc_create_ingress(pmd->nlsk_fd,
1360 pmd->remote_if_index) < 0) {
1361 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1365 LIST_INIT(&pmd->implicit_flows);
1366 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1367 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1368 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1369 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1371 "%s: failed to create implicit rules.\n",
1380 RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1381 strerror(errno), errno);
1382 if (strlen(remote_iface)) {
1383 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1389 RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1390 strerror(errno), errno);
1391 tap_flow_implicit_flush(pmd, NULL);
1394 RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1395 rte_vdev_device_name(vdev));
1402 set_interface_name(const char *key __rte_unused,
1406 char *name = (char *)extra_args;
1409 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1411 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1412 DEFAULT_TAP_NAME, (tap_unit - 1));
1418 set_interface_speed(const char *key __rte_unused,
1422 *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1428 set_remote_iface(const char *key __rte_unused,
1432 char *name = (char *)extra_args;
1435 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1441 set_mac_type(const char *key __rte_unused,
1446 !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1447 *(int *)extra_args = 1;
1451 /* Open a TAP interface device.
1454 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1456 const char *name, *params;
1458 struct rte_kvargs *kvlist = NULL;
1460 char tap_name[RTE_ETH_NAME_MAX_LEN];
1461 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1462 int fixed_mac_type = 0;
1464 name = rte_vdev_device_name(dev);
1465 params = rte_vdev_device_args(dev);
1467 speed = ETH_SPEED_NUM_10G;
1468 snprintf(tap_name, sizeof(tap_name), "%s%d",
1469 DEFAULT_TAP_NAME, tap_unit++);
1470 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1472 if (params && (params[0] != '\0')) {
1473 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1475 kvlist = rte_kvargs_parse(params, valid_arguments);
1477 if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1478 ret = rte_kvargs_process(kvlist,
1480 &set_interface_speed,
1486 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1487 ret = rte_kvargs_process(kvlist,
1489 &set_interface_name,
1495 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1496 ret = rte_kvargs_process(kvlist,
1504 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1505 ret = rte_kvargs_process(kvlist,
1514 pmd_link.link_speed = speed;
1516 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1519 ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1523 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1525 tap_unit--; /* Restore the unit number */
1527 rte_kvargs_free(kvlist);
1532 /* detach a TAP device.
1535 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1537 struct rte_eth_dev *eth_dev = NULL;
1538 struct pmd_internals *internals;
1541 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1544 /* find the ethdev entry */
1545 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1549 internals = eth_dev->data->dev_private;
1550 if (internals->nlsk_fd) {
1551 tap_flow_flush(eth_dev, NULL);
1552 tap_flow_implicit_flush(internals, NULL);
1553 nl_final(internals->nlsk_fd);
1555 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1556 if (internals->rxq[i].fd != -1) {
1557 close(internals->rxq[i].fd);
1558 internals->rxq[i].fd = -1;
1560 if (internals->txq[i].fd != -1) {
1561 close(internals->txq[i].fd);
1562 internals->txq[i].fd = -1;
1566 close(internals->ioctl_sock);
1567 rte_free(eth_dev->data->dev_private);
1568 rte_free(eth_dev->data);
1570 rte_eth_dev_release_port(eth_dev);
1575 static struct rte_vdev_driver pmd_tap_drv = {
1576 .probe = rte_pmd_tap_probe,
1577 .remove = rte_pmd_tap_remove,
1579 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1580 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1581 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1582 ETH_TAP_IFACE_ARG "=<string> "
1583 ETH_TAP_SPEED_ARG "=<int> "
1584 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1585 ETH_TAP_REMOTE_ARG "=<string>");